Resnet pretrained weights and tips for high resolution inputs

[zzzrenn]

Thank you for your great work. It looks promising and I would like to try it out with high resolution inputs. Unfortunately, the inference speed is not up to expectation for high resolution input (1280x1280). I plan to replace the ViT with ResNet to avoid expensive attention computation. Could you please release the pretrained weights for Resnets and share tips on dealing with high resolution inputs?